The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill processes account names and domains from the CRM model, which are interpolated into the agent's prompt. This represents a potential indirect prompt injection surface where adversarial content in a company name could attempt to influence the agent's logic.
Ingestion points: Untrusted data enters the prompt in infra/plays/score-accounts.ts via the input object containing the account name and domain.
Boundary markers: The prompt does not utilize specific delimiters to isolate the account name and domain strings within the agent instruction.
Capability inventory: The skill possesses capabilities to write to the CRM via hubspot.updateRecords in infra/plays/score-accounts.ts and perform business data enrichment in infra/agents/scorer.ts.
Sanitization: No sanitization of the account name or domain is performed prior to prompt interpolation.
[EXTERNAL_DOWNLOADS]: The skill references the @cargo-ai/cli package in its installation metadata (SKILL.md). This is the standard command-line utility provided by the platform vendor and is used for legitimate skill deployment and workspace management.

account-scoring